2.1 A simple agency problem

There are a great many interactions in the labor market that can be fruitfully examined as a principal agent problem—an interaction in which the principal uses a reward structure to motivate an agent to pursue some desired action. As a baseline example, consider a principal who seeks to maximize profit, which depends on the “effort” of an agent and the compensation given to that agent, as specified by

     (1)

In (1), is the value produced for the principal as a consequence of an agent’s effort  (which can be represented as a non-negative scalar) and is compensation given to the agent. We assume that and exist and are continuous. As for the agent, we assume simply that utility is . Thus is the money metric disutility of taking the action that benefits the principal. What makes this problem interesting is that the principal cannot directly observe effort.

Although information asymmetry is essential to our story, to set basic ideas, we ask initially what the solution would be if the principal could observe the agent’s effort, and write a contract specifying effort and wage. The firm would then simply maximize (1), subject to the agent’s participation constraint, which specifies that the utility resulting from the agreed-upon wage and effort must equal or exceed the utility available to the agent elsewhere, i.e., . The principal finds it most profitable to operate with this latter constraint binding, so it immediately follows that the solution to this constrained maximization problem entails

     (2)

This outcome is efficient: the marginal value of additional effort equals the marginal cost to the worker of supplying the effort. This agreed-upon effort level is the same as if the agent worked for herself. The resulting wage is .

When, instead, the principal does not observe the agent’s effort level, the principal must find an incentive mechanism to induce the desired effort level. One possibility is that pay can be conditioned on the value of output . In some instances, though, neither output nor effort are readily observable. We consider such a case. We suppose, instead, that the principal has a noisy signal of effort,

     (3)

where is drawn from a differentiable, symmetric, single-peaked density (with corresponding cumulative density ). There are variety of possible interpretations consistent with this set-up. For example, might be some objective measure of performance, and is simply luck. Alternatively, we might interpret as a principal’s impression or opinion of how well a worker is performing, so is unobservable by outside parties. The term in this latter case captures miscommunication and misinterpretation of effort.

We start with the case in which the realized value of is common knowledge, adopting the assumption that the principal can be trusted to honor commitments in which compensation is conditioned on . In this case, there are many incentive schemes that will do. To set the stage for results to come, we work through a particularly simple scheme: we assume that the principal commits to a policy of paying wage if the observed performance falls below some cut-off , and paying if is above that cut-off.¹

To be clear, we have the following timing in mind: (i) The principal announces the policy (including ), and posts and . (ii) The agent decides whether to accept the job, and if she does, takes hidden action . (iii) Nature plays . (iv) Given , the firm pays the agreed-upon wage.

We can easily solve for the optimal wage policy, . Conceptually, the first step is to account for the agent’s best response to the wage policy. At effort level , the probability of earning is and the probability of earning is . So the agent wants to maximize

     (4)

which leads to the best response, , that solves

     (5)

From this last expression, we notice that the best response is a function of the difference between the higher wage and the lower wage, say ( is the “bonus” that accompanies the high-performance outcome).² Thus we can write , noting, for future reference that

     (6)

(under the assumption that the second order condition holds). This makes sense; higher-powered incentives elicit greater effort.

Next, the principal must account for a participation constraint. The agent accepts the job only if the expected wage equals or exceeds the agent’s opportunity cost:

     (7)

The principal’s problem then turns out to be straightforward. Expected profit is output minus the expected wage, and given that the participation constraint binds, this is just

     (8)

The first order condition to the principal’s profit maximization problem is

     (9)

Above, we noted that for any best response, so the elicited effort level, , described by (9) solves

     (10)

The solution thus entails the efficient level of effort, as in (2). The principal pins down using (5), which can be read as giving as an implicit function of . Finally, given and , the firm sets the base wage to be as low as possible, while still meeting the participation constraint (7).

We have obviously chosen to work out an extremely simple case as our prototypical principal-agent model.³ As simple as the model is, it is nevertheless sufficient to make the point that a solution to the agency problem entails a strategy of conditioning pay on observed productivity. This reward structure can elicit efficient effort levels even when agents are entirely self-interested and when performance measures are noisy and imperfect.

2.2 Agency matters

In this section we demonstrate the value of thinking carefully about agency in the context of three labor market applications: (i) CEO compensation, a case in which there is a single agent, (ii) personnel policies in a firm, which involves a single principal seeking to motivate multiple agents, and (iii) unemployment and labor market segmentation that can arise in labor markets in which multiple principals compete for agents, and in which the motivation problem is addressed by the threat of dismissal. In each of these three cases, solutions to the principal agent problem are seen to have important social consequences. In each case also, empirical evidence indicates that anomalies exist that point to the importance of behavioral aspects that are not included in the standard principal agent set-up.

2.2.3 Involuntary unemployment and market segmentation

In 1984, Shapiro and Stiglitz set out an influential “efficiency wage” model that illustrates an important feature of agency models: the actions an individual firm takes to resolve an agency problem can give rise to important social costs when adopted throughout the market. In the case of efficiency wage models, the social costs are those arising from involuntary unemployment and labor market segmentation.¹⁵ The set-up we present here is a recasting of the Shapiro and Stiglitz model taken from Ritter and Taylor (forthcoming).

We consider a market in which there are a large number of identical profit-maximizing employers, each of which faces the agency problem we outlined above. Each firm in the model is assumed to behave as outlined in Section 2.1: the idea is to pay well for “good outcomes” while penalizing workers for “bad outcomes” to the maximum extent possible. Limited liability is invoked through the assumption that the only penalty that the firm can implement is to dismiss a poorly performing worker. The motivation problem is resolved by employers making jobs sufficiently valuable that workers will provide effort so as to prevent dismissal.

To capture the idea that jobs have value, it is necessary to set the model up in a multi-period framework. The agency model outlined in Section 2.1 is thus assumed to pertain for each period indefinitely and workers are assumed to live indefinitely with a discount rate .¹⁶

The basic set-up

Employees are paid for one unit of labor per period. In each period a worker chooses , and this produces utility . In this model, the alternative to employment is unemployment, which results in utility in the period. The present value of being unemployed is (which is not 0 because there is some prospect of being hired in the future). Hiring and termination are costless to the firm.

The model is a game between the firm and a worker with the following order of play in each round: (1) The firm offers a wage . (2) The worker chooses effort level . (3) Nature plays using the distribution . (4) The firm pays . (5) The firm decides whether to retain the worker or end the game. We focus on the perfect Bayesian equilibrium in which the worker is retained if and only if exceeds an endogenous threshold . We assume that is common knowledge. (Workers can infer by observing the frequency of terminations.)

The solution method mimics the steps we took in the simpler model above. In particular, we first find the worker’s best response. Then we see how the firm will chose its personnel policies ( and ) in light of the worker’s best response.

The worker’s best response

Let be the worker’s best response. To find that best response notice that for a person who chooses in the current period, and then reverts to in all future periods, lifetime utility is given by

     (16)

The employee maximizes by choosing . For an interior solution, the first order condition is

     (17)

As in our baseline agency model, the second order condition holds when . As we have noted, this incentive elicits effort because the job is valuable: . As is typical of models that invoke limited liability, the participation constraint does not bind.

Evaluating Eq. (16) at and solving for , then substituting into the first order condition (17) produces

     (18)

This last expression implicitly defines the worker’s best response, .

Firm profit maximization

Now we can turn to the firm’s objective. It seeks to maximize profits, taking into account the worker’s best response, i.e., maximizes

The solution can readily be found using the same steps we followed in solving the agency problem in Section 2.1. In this instance the optimal employment policy again induces the socially optimal performance level regardless of :¹⁷

The noise in the environment does, however, affect the distribution of surplus. In particular, Ritter and Taylor (forthcoming) establish the following results: First, when the firm optimally chooses and , the resulting probability of retention, , is unaffected by the level of (the variance of the density ).¹⁸ Second, the optimal wage does depend on , as follows:

     (19)

where is a “standardized” random variable, , and is the “standardized p.d.f.,” i.e., the p.d.f. of . The more intractable the agency problem—the greater the value of —the higher is the wage required to achieve efficient effort and so the greater the surplus accruing to the worker.

Equilibrium unemployment

The equilibrium of the model we have just outlined generates unemployment.

Let be the present value of lifetime utility of an employed individual who works at the optimal effort level. Recall that is the expected lifetime utility for an individual who is unemployed. This utility level depends, clearly, on the probability of job acquisition. Let that rate be .¹⁹ Given that current-period utility of an unemployment person is zero, the expected lifetime utility of an unemployed individual is

     (20)

so in turn we can use (17) and (20) to solve for and substitute into Eq. (19), giving

     (21)

Equation (21) gives the locus of potential equilibrium values of and .

Now job loss among the employed occurs with probability , while job finding among the unemployed occurs with probability . So if is the steady state unemployment rate, we must have . Solving for and substituting into (21), and substituting also for using Footnote 18, gives

     (22)

This expression shows potential equilibrium wage and unemployment levels for the labor market.

Figure 1 shows the market equilibrium when long-run labor demand is perfectly elastic, at , with indicating available workers and indicating employment. (More general formulations are easily handled.) Equilibrium unemployment, , solves

Figure 1 Equilibrium wage and employment.

     (23)

Clearly . Also, inspection of (23) shows that an increase in increases unemployment. This outcome is intuitive. The weaker the link between the dismissal threat and employee behaviors, the stronger are the incentives required to elicit the desired effort level. In equilibrium, heightened incentives require higher unemployment.

This model of equilibrium unemployment—the Shapiro-Stiglitz model—has emerged as a workhorse for the analysis of macro-labor issues. It has proved to be useful also for evaluating policies like unemployment benefits, the public interest in regulating firms’ layoff decisions (i.e., just-cause dismissal requirements, as discussed in Levine, 1991), and the potential of minimum wage policy to actually increase employment (Rebitzer and Taylor, 1995b). Having said this, economists are divided on the extent to which efficiency wages are an important source of equilibrium unemployment. Other forces, like labor market frictions, matter as well in determining equilibrium unemployment rates.²⁰ Efficiency wages are clearly not the whole story.

Labor market segmentation

Although we have focused on unemployment, the lost output from unemployment may not capture the full social costs of efficiency wage personnel policies. After all, if there are some jobs in the labor market where agency issues are of little importance, workers should generally be able to find jobs there. From an efficiency perspective, finding work in these “secondary jobs” is similar to unemployment in that individuals in the secondary labor market would prefer higher-productivity “primary jobs,” but the equilibrium supply of qualified workers for these jobs exceeds the demand.

Labor market segmentation emerges if we enrich our efficiency wage model by allowing the difficulty of agency problems to vary across firms. Recall from (21), that in the Ritter-Taylor version of the Shapiro-Stiglitz model firms choose to pay

     (24)

where and are positive constants that are independent of . Firms that have low values of , i.e., who have production processes with accurate measures of worker effort, can pay wages that are relatively low. On the other hand, firms will choose to set wages high when effort is hard monitor or, equivalently, when they face high values of .

This latter observation was emphasized in Bulow and Summers’ (1986) paper on “dual labor markets.” In their conception, firms with severe agency problems pay high wages and are said to belong in the primary sector. The strategy of paying high wages is effective because workers are motivated by the prospect of retaining valuable jobs. Thus we also expect to observe low levels of voluntary exit from such firms and efforts on the part of firms to retain workers even in a down-turn. In contrast to the primary sector, firms that have modest agency problems can pay wages that are close to the market-clearing level. These secondary sector firms will be less concerned about worker turnover. In an extension of this argument (Rebitzer and Taylor, 1991) we show that firms which employ efficiency wages as a motivating device will also be led to hoard labor, i.e., employ labor above the value-of-marginal-product curve. By taking actions to ensure future employment—perhaps by hiring contingent workers to absorb demand shocks—firms can reduce the wage needed to provide optimal motivation to workers.

The most widely examined empirical prediction of efficiency wage models of labor market segmentation is that there will be cross-firm and cross-industry wage variation resulting from firm characteristics, rather than worker characteristics. There is considerable evidence for industry and firm wage effects (including well-known work by Krueger and Summers (1988), on industry effects, and Brown and Medoff (1989), on firm size effects) but it is often unclear how much this is due to factors emphasized in efficiency wage models (such as monitoring difficulties) or other market imperfections such as those emerging from search frictions (e.g., Burdett and Mortensen, 1998).

One potentially helpful approach entails the study of specific firms and industries, with an eye toward the central predictions of the model. Thus, Cappelli and Chauvin (1991) examined worker performance across plants within the same firm, examining the extent to which workers seem to choose performance on the basis of the value of their job relative to other opportunities in their local labor market. They find evidence that is generally supportive of the efficiency wage set-up. Similarly, in work with Daniel Nagin and Seth Sanders, (Nagin et al., 2002), we evaluated a field experiment in which a firm manipulated monitoring rates across several work sites. Consistent with the effort-regulation model set up above, there was substantially more malfeasance in locations with low monitoring levels.

At the broadest level, the efficiency wage literature points to important social costs that emerge as a result of the strategies individual firms use to resolve agency problems. If firms indeed rely on the fear of job loss to motivate employees, labor markets can be expected to waste human capital on a large scale through involuntary unemployment and labor market segmentation. If, however, other motivators can be mobilized to resolve agency problems, the situation may not be so grim. The costs of agency problems might be further reduced if schools can socialize children to be especially responsive to these alternative motivators. Indeed, some have speculated that such socialization may be the source of much of the social and private returns to investments in human capital. We take up some of these alternative motivators in Section 4 below. Before turning to these, however, we must first introduce another conceptual building block that is important for our story—incentives that are intended to work along more than one dimension.

3.1 High wages as a signal of firm fitness

We begin by discussing a theoretical issue that is well known in the literature on efficiency wages. As we have seen, firms that pay efficiency wages must set wages above the market clearing level to elicit the desired level of effort. Effort can, however, be elicited more cheaply by a deferred compensation policy that causes employees to, in effect, post a performance bond. By judiciously back-loading pay, firms can create powerful work incentives while choosing a wage path whose discounted present value is equal to the market clearing wage. With this option available, why would employers ever select a more costly efficiency wage strategy? The practical relevance of this theoretical puzzle is sharpened by empirical work suggesting that even when very large amounts of deferred compensation were available, as is the case in the promotion from associate to partner in large corporate law firms, firms set wages as if they were pursuing an efficiency wage strategy (Rebitzer and Taylor, 1995a).

Ritter and Taylor (1994) tackled this issue by observing that for both efficiency wages and the performance-bond incentive, the power to shape behavior depends on the likelihood that the firm will honor its future commitments to employees. All else equal, firms will more effectively solve agency problems if employees expect them to be highly reliable in honoring future wage commitments.

Ritter and Taylor build upon this insight by positing a market in which there are two types of firms: highly reliable firms (i.e., firms that are unlikely to go bankrupt or otherwise renege on commitments) and less reliable firms (firms that are more likely to become bankrupt or renege). Reliability is known by the firms but not by anyone else, though the distribution of types is common knowledge. Firms would like to resolve their agency problem as cheaply as possible, and are inclined to do so by asking workers to post bonds in the form of deferred compensation. The posted bond is forfeited if the worker is judged to be working at a sub-standard effort level but is returned, with interest, if the worker’s observed performance meets the expected standard.

Under some conditions, all firms pursue the same deferred compensation strategy. In this pooling equilibrium, workers will require a rate of return on their bonds that reflects the aggregate level of riskiness, based on the market-wide probability a firm will fail and be unable to return the bond.²¹ This, of course, is a good deal for low-reliability firms—who benefit by paying a below-market interest rate on the bonds that workers have posted—but a bad deal for high-reliability firms. A more interesting possibility is that efficiency wage strategies might emerge for some firms as a separating equilibrium. In this equilibrium, a reliable firm that deviates from the bonding strategy—by paying a high wage up front—offers a credible signal that it is a highly desirable counterparty for long-term relationships. If the offered wage is sufficiently high, low-reliability firms will find it unprofitable to mimic this strategy, and the equilibrium thus satisfies the “Intuitive Criterion” of Cho and Kreps (1987). Equilibrium efficiency wages arise endogenously, in short, without a recourse to limited liability arguments.²²

Our primary point here concerns the use of incentives policies to do “double duty.” In the separating equilibrium, highly reliable firms use wage policies to solve an agency problem and to signal the fitness of the firm. In order to pursue both objectives, these firms must compromise on their use of deferred compensation and this compromise necessarily introduces distortion. Thus, in the separating equilibrium there is a surplus transfer to workers employed by highly reliable firms and employment in the reliable-firm sector is inefficiently low.

3.2 The rat race

Rat race models build on a simple observation: early in their careers many successful professionals appear to be overworking. It is commonplace to find lawyers, consultants, and assistant professors complaining that the hours they work are simply “too much” and that they interfere with forming and raising a family. These strains are increased by the dramatic influx of women into professional occupations because overwork is most intense during prime years for family formation and childbearing. From the point of view of simple models of labor markets, this sort of overwork is anomalous. Firms are in competition for talent, and it would seem that the most successful competitors would be those who best accommodate employee preferences about work conditions—including work hours. In his famous paper on the “rat race,” Akerlof (1976) offers a potential resolution to this anomaly based on unobservable worker heterogeneity.

Akerlof’s set-up focuses on a production line. At the end of the day, line workers are jointly rewarded on the basis of total output. There are two types of workers—those inclined to work hard and those inclined to work less hard. To employers these workers appear identical, so they all earn the same wage. If both types of workers accept positions on the production line, this is a great deal for low-effort workers (who would earn lower pay than high-effort workers in a perfect-information world) but a bad deal for high-effort workers. This is precisely the situation that might lead firms to adopt rules that will provide high-effort workers the opportunity to credibly signal that they are in fact high-effort workers. Thus, Akerlof’s proposed solution is that the firm set the production line at a speed that is uncomfortably fast for high-effort workers but more uncomfortable yet for low-effort workers—so uncomfortable indeed that the low-effort workers will opt out of working for the firm. The rat race thereby serves the useful function of screening out the low-effort workers.

In Akerlof’s model, compensation policies are, quite clearly, doing “double duty.” Compensation arrangements and work conditions are structured (i) to compensate workers at a level necessary to induce them to accept employment at the firm, and (ii) to create incentives that attract the “right kind” of worker to the firm. The distortion here is that workers are being asking to provide effort that exceeds the first-best level. In a marketplace with many employers, the market can devolve into an equilibrium in which all firms that hire high-effort workers ask those workers to work at uncomfortable effort levels. This “adverse selection” equilibrium occurs because any one firm failing to adopt a rat race would be swamped by low-effort workers. The equilibrium is inefficient, in the sense that all firms would experience increased profitability if they coordinated on a lower-effort work norm.²³

Akerlof’s demonstration of an overwork equilibrium was presented in a self-consciously unrealistic example, but subsequent theoretical and empirical work suggests that it is an important phenomenon in professional labor markets. For instance, in a paper with Renee Landers (Landers et al., 1996), we embed Akerlof’s idea into a simple tournament-partnership model designed to shed light on work practices in large US law firms. In our two-period model, young lawyers accept salaried positions as “associates” for one period, and if deemed suitable are promoted to be “partners” (equity shareholders) in the subsequent period. Partners share equally in firm surplus. This equal sharing rule gives incumbent partners powerful incentives to promote only highly motivated lawyers into the partnership.

We assume that there are two types of lawyers who are equally productive but have differing preferences over the hours they prefer to work: these are “short-hour” and “long-hour” workers. Now in our setting, firms have the incentive to attract workers who will be inclined to work long hours. The reason is that when workers become partners—at which point they share firm surplus—the long-hours individuals will engage in less free riding. As in Akerlof’s model, an adverse selection equilibrium emerges. In our case, associates’ willingness to put in extended hours over many years serves as a credible signal that they are long-hour individuals.

Empirical evidence for the relevance of the over-work equilibrium comes from an empirical examination of work hours and work preferences in associates at two large East Coast law firms. In a survey conducted in these firms, we find that most associates express a preference for working shorter hours (with a correspondingly lower salary) but, importantly, their willingness to work shorter hours hinges on the work-hour norms adopted by other associates. In addition, we find that in making promotion decisions, partners use an associate’s willingness to work long hours as an indicator of the motivation associates have to excel. These findings would not be expected in a conventional labor market, but are precisely what one would expect if work hours are being used as a signal of one’s otherwise-unobservable type.

In our paper (Landers et al., 1996) we abstract from “career concerns” outside one’s own firm, but it is clear that overwork early in one’s career might be valuable not only as a signal within a firm but as a means of career advancement elsewhere. Completion of six years as an associate at a law firm known for abusing associates with grueling hours can be a valuable means of demonstrating an important but hard-to-observe trait to other employers in the marketplace.

The key idea that current on-the-job behaviors can affect one’s future career, through their impact on reputation, is studied in insightful papers by Holmström (1999) and Gibbons and Waldman (1999). Gicheva (2009) shows how long work hours early in one’s career can affect an individual’s value in the market later in the career. Gicheva shows, further, that her model helps explain wage growth in a sample of workers who took the Graduate Management Admissions Test (GMAT). Specifically, she shows that among workers who worked above-norm hours when they were young (48 or more hours per week), subsequent wage growth was positively correlated with early career hours worked. The same was not true for those workers who worked fewer than 48 hours; for those workers wage growth was uncorrelated with hours worked.

Signalling can be particularly dysfunctional in situations in which workers can devote effort to more than one task—an issue we take up next—because it can distort effort allocation. An example is given in the work of Acemoglu et al. (2008). In their model, career concerns can motivate excessive and misguided signaling by primary school teachers—misguided because some effort is devoted to improving the signal (student performance on a proficiency test), without actually improving students’ true human capital. An important conclusion in that paper is that the problem of excessive signalling can fundamentally shape the desirability of using markets versus the government for the provision of some services. Specifically, in a competitive market, there will be socially costly distortions as the result of excessive signalling. An advantage of having teachers employed in the public sector is that the government might be able to commit to policies that eliminate excessive signalling.

3.3 Multi-tasking

Perhaps the most obvious case of dual-purpose incentives occurs when a principal seeks to regulate an agent’s behaviors along more than one dimension. We have already encountered examples along these lines in our discussions above. For instance, in our examination of CEO compensation, we noted problems that arise when a compensation board seeks to create incentives for a CEO to exert effort toward increasing shareholder value and limiting wasteful expenditure in the executive suite. Our discussion of tournament incentives suggests a second obvious example: What happens in a tournament when each worker must be motivated to provide effort along his own assigned task and be motivated also to be cooperative with other workers?²⁴ A third example is Acemoglu, Kremer, and Mian’s work, mentioned in the previous paragraph, in which teachers allocate effort that improves student human capital and effort that merely improves a student’s test score (“teaching to the test”).

Holmström and Milgrom (1991) establish a number of insightful and surprising results in precisely such contexts. The central point of their paper is both simple and profound: when an agent performs multiple tasks, incentives must perform the double duty of inducing appropriately high levels of effort generally and inducing a desirable allocation of an agent’s attention across the various tasks inherent in the job.

We can get a feel for their analysis by making a simple extension to our baseline principal agent model. Let us suppose now that the agent can allocate effort along two dimensions, and . We suppose also that the agent’s utility is now , and, following Holmström and Milgrom, we make two key assumption about the function that gives money metric disutility of effort, : (i) it is convex and (ii) it achieves a maximum at a positive level of effort. This last assumption means that in the absence of direct incentives the worker will be happiest when putting forth some effort. The principal’s objective continues to be the maximization of value added by the worker, now given by .

A principal who has reasonably accurate measures of and , say and , will be able to construct an incentive scheme in which bonuses reward each dimension of effort appropriately. Matters are more interesting when the principal has good information along one dimension of effort but not the other. To take an extreme example, suppose the firm has a subjective measure , but no measure at all of . The firm’s best strategy then will depend crucially on the nature of the production function, . To see how this works, we set to be , with representing the agent’s “bliss” level of effort, and work out the optimal incentives for two different production functions: first, a case in which the two types of effort are perfect substitutes, ; second, a case in which they are complements, .

The solution to the first case is easy to characterize. If the firm places any incentive whatsoever on the first type of effort (i.e., a bonus based on ), the worker’s best response will be , and with this in mind the firm can follow the steps outlined for our baseline one-dimensional principal agent model. It is easy to confirm that the result will be to elicit effort . Intuitively, the firm will prefer this strategy if the value of the first type of effort is high relative to the second type of effort, which is certainly true if . On the other hand, if is sufficiently large, the principal might decide to use no explicit incentive and instead ask (nicely!) that the worker direct all his effort along the second dimension. Given a binding participation constraint, , it is easily shown that value added by the worker is

     (25)

As anticipated, if is sufficiently high relative to , the firm will simply place incentives on the observable portion of performance. This is efficient when , and is second-best optimal even when is moderately lower than . However, when is sufficiently low relative to , the firm will instead try a “cooperative” strategy. No incentives are used; the worker is simply asked to direct all effort to the second dimension.

The solution to the second case, in which the two types of effort are complements, is also intuitive. When , the firm clearly must avoid a best response of , and so will never use an explicit incentive along the first effort dimension. In this case the firm instead directs the worker effort to be and hopes that the worker complies.

Simple as this analysis is, several interesting points emerge:

First, we see that there will be cases in which an employer will choose not to place an incentive on an easily-observed dimension of performance, even when that effort is valuable to the firm. This happens when there is a similarly-valuable second dimension of effort that is sufficiently difficult to observe and incentivize. Such an outcome is particularly likely when multiple tasks are complementary. In such cases the firm is best off using very low-powered incentives, i.e., simply paying a base wage.

Second, as shown by our second example, the principal’s optimal incentive plan can result in a second-best outcome that is far from efficient. For instance, when , it is easily confirmed that the efficient level of output is .²⁵ The firm’s low-powered incentive scheme results, instead, in output . If possible the firm would very much like to find a way to make this worker a residual claimant, and indeed would be willing to suffer substantial cost along some other dimension to make this happen. In short, a strong motive exists here to outsource the task at hand to an independent contractor if this can be made workable.²⁶

Third, when a firm cannot use independent contracting, or for some compelling reason chooses not to, a central goal of the firm often entails structuring activities to improve management’s ability to closely monitor and supervise key contributions by employees. Put another way, the issue of multi-tasking can matter a great deal for the organization of firm production.

Finally, and most important for our purposes, the multi-tasking approach developed by Holmstrm and Milgrom, and illustrated with the simple example above, clearly opens the door for “behavioral factors” to play a central role. For example, in our model, the “bliss” level of effort is taken to be an exogenous constant. This unfortunate abstraction overlooks the many sociological and psychological factors that determine how intrinsically motivated individuals contribute to the success of their firm. These are the sorts of consideration that lead one into the territory of behavioral economics.

4.1 Pay status: Financial incentives and inequality aversion within firms

People dislike inequality—especially when they have drawn the short straw. Indeed, there is substantial indirect evidence that wellbeing is shaped in large measure by comparisons with others.³⁰ This idea can matter within organizations because people are likely to compare themselves with others around them in the workplace. In turn this can be an important determinant in shaping firm compensation policies.

The idea that interpersonal comparisons matter to agents can easily be captured by including “asymmetric inequality aversion” into utility functions. Utility is increasing in one’s own income, of course, but decreasing in the income of other relatively-wealthier comparison individuals. The “asymmetry” refers to a presumption that agents suffer more from inequality that is to their material disadvantage than they gain from inequality that is to their material advantage (see, e.g., Fehr and Schmidt, 1999).³¹

As an example of how asymmetric inequality aversion can affect our agency models, recall the tournament model, as set out in Section 2.2.2. In that model, workers had utility given by , and made a net contribution of . The firm conditioned pay on an imperfect measure of : it paid a base wage to all workers, and in addition gave a bonus to the fraction of workers who had the highest observed performance. (The “bonus” in this case would typically be a promotion to a higher-paying job.) Given workers’ best-response effort choices, we saw that this simple tournament resulting in all workers supplying the efficient effort level, i.e., they choose that solves .

Suppose we take that same model but now introduce asymmetric inequality aversion by letting

     (26)

Here, reflects the possibility that winners feel empathy for losers, proportional to the inequality generated (but of course people do like to win, so ). reflects the fact that workers who do not win the bonus suffer an even large utility loss due to inequality aversion. Repeating steps outlined in Section 2.2.2, we can show that the principal’s solution now has a first-order condition

     (27)

and, since , we have , which in turn means that the firm here settles for a second-best effort level, .

Inequality aversion causes the incentive pay parameter to do the “double duty” of eliciting work effort and determining the extent of expected pay inequality in the firm. As a result, the firm must compromise along an important dimension by lowering incentive pay, , and reducing the effort level elicited from workers. Equation (27) also shows that in this setting the firm will want to be careful how it sets its promotion rate. Here the firm would like (all else equal) to set near 0, which would allow effort to approach first-best. Intuitively, the cost to the firm of inequality is lowest when there are relatively few people who are affected by the inequality, i.e., when the promotion rate, , is close to 1.

The logic of this model of income comparisons underlies Frank’s (1984) seminal article on inequality aversion in labor markets. In his treatment, an employee gains in utility from being high in the firm’s pay hierarchy and loses utility from having a low position. Just as in our model, these concerns cause firms to operate with lower-powered incentives. Frank presents evidence drawn from many different types of organizations; he sees, for example, a dampening effect in commissions paid to car salesman and realtors as well as pay compression among college professors.

Encinosa et al. (2007) present a similar analysis of these issues in the context of incentive pay within medical partnerships. In these professional organizations, physicians determine incentive intensity by choosing how broadly they wish to share the income they generate with others in the practice. For example, groups often choose to share income equally across physicians—in a practice with physicians, each physician keeps of profits—which minimizes inequality. The practice of equal sharing rules has the potential disadvantage, though, of offering the lowest possible level of incentive intensity to partners. The model of inequality aversion set out by Encinosa, et al. shows that the tension between these forces makes sharing rules less attractive in large partnerships than in smaller partnerships—a result supported by the available data. The authors also present evidence consistent the notion that physicians compare effort as well as income. We take up effort comparisons in the next section.

As a second example of the potential impact of inequality aversion in a principal agent setting, consider the multi-tasking model we examined in Section 3.3. Recall that in that model, we assumed an agent’s utility is represented by , where is a positive constant, and we assumed further that the principal had a good signal for but not . So when the principal’s payoff is , the best the principal can do is pay the agent a fixed wage that meets the participation constraint (i.e., so that ) and then direct the agent to allocate effort so that . Now suppose that the agent is inequality averse, just as in (26), but that in this case her comparison is the principal’s income, . So for the agent, utility is

     (28)

where reflects the extent to which the agent is inequality averse. The important point here is that the agent can always costlessly enforce perfect equality by simply adjusting effort allocation (keeping total effort at ); she can reduce the principal’s profits, while causing no harm to herself. If the principal sets the wage to , the agent will adjust effort so that also equals . So the principal typically finds it profitable to increase above the participation constraint, i.e., the principal uses “rent sharing.” Reducing wages to the level of the employee’s outside option would be self-defeating in this context because it risks that the agent will become disgruntled and take steps to “even the score.”

The idea that inequity aversion supports rent sharing has been extensively explored in laboratory and field experiments involving variations on the Ultimatum Game. In this bargaining game a proposer offers to divide a fixed amount of money between himself and a responder. The responder can accept or reject this offer. If the offer is accepted, the money is divided according to the offer. If the offer is rejected by the responder, however, neither the responder nor the proposer gets any money. The conventional game theoretic solution to this bargaining problem for selfish players is for the proposer to make an offer in which he keeps all (or nearly all) of the money while the responder accepts any offer.

It turns out, however, that the participants in these games don’t behave as entirely selfish players. Proposers routinely make offers close to an equal division of the pie and responders routinely reject low offers. These anomalies can be resolved, of course, by introducing equity concerns into players’ utility functions. Inequality averse proposers get less utility than entirely selfish players do by proposing a division of the pie that greatly favors themselves. Conversely, inequality averse responders can credibly threaten to destroy the surplus if highly unequal divisions are proposed. In practice, reasonably egalitarian offers are made and accepted.

Most employment relationships exist in the context of labor markets. Thus it is not sufficient to demonstrate that individuals prefer more equitable pay practices. Economists must also establish that these preferences matter for the equilibria that emerge in markets. Fehr and Schmidt (1999) examine this central issue by considering whether the egalitarian rent sharing observed in the Ultimatum Game survives in an environment in which there are many proposers and a single responder who must accept or reject the best offer received. They find that competition between proposers leads to lower levels of egalitarianism. To see the logic, consider a situation in which proposers each offered 50 percent to the responder, leaving each proposer with a chance of having his offer accepted. An individual proposer could clearly do better by simply offering a 51 percent share to the responder, thereby insuring that his offer was selected (the probability increases from to 1). But all proposers are driven by this same logic, and in the end the responder gets all the surplus. This, of course, is the outcome we would observe if proposers had no inequality aversion.

The irrelevance of inequality aversion stems from the fact that with many competitors, no single player can prevent an inequitable outcome. If no individual action can reduce the inequality of the equilibrium outcome, then inequality aversion cannot be an important determinant of behavior. Fehr and Schmidt conclude that matters are different when individual players have a way to impose a cost on the counterparty to a highly unequal offer (as does the agent in the example we consider with Eq. (28)). Specifically,

In short, agency problems of the sort depicted in our model above (that allows retaliation motivated by equity concerns), can survive market competition.

The Fehr-Schmidt conjecture has been examined experimentally by Fischbacher et al. (2009).³² They find that increasing proposer competition in the Ultimatum Game, by adding extra proposers, causes a large increase in mean accepted offers. Similarly, increasing responder competition causes a reduction in mean accepted offers. Both of these results suggest that competition undercuts the influence of equity norms on bargaining outcomes, although in each case the increase in inequality of outcomes is less than one would predict on the basis of competition alone.

The idea that workers exact retribution upon employers when treated unfairly is supported by Bewley’s (1999) extensive qualitative interviews. He found that managers and other labor market participants believe that there is a connection between employee morale and performance. Bewley focuses on the morale effects of cutting wages in recessions: Employers are averse to cutting wages because of the fear of a backlash from employees.

Evidence that the fear of backlash is reasonable emerges from a number of recent studies by Mas and co-authors. In a remarkable study, Mas (2006) finds that when New Jersey police officers lose in final offer arbitration, so that the wage they receive is lower than the requested wage, arrest rates and average sentence lengths decline, while crime reports rise. This evidence is consistent with the idea that workers are less inclined to provide effort when the wage falls below a salient reference wage. Krueger and Mas (2004) report evidence that a long and contentious strike and the hiring of replacement workers in a Bridgestone/Firestone plant contributed to the production of defective tires. Mas (2008) finds Caterpillar plants that underwent contract disputes experienced reduced workmanship and reduced product quality. In this latter study he estimates that the contract dispute was associated with at least $400 million in lost service flows due to inferior quality.

The papers cited in the previous paragraph study workers in unionized environments. The presence of a union likely facilitates the sort of collective retaliation that punishes employers who take morale lowering actions. The central idea, that perceptions of unfairness can damage effort, is likely to be important outside the union sector. Evidence along these lines appears, for example, in our field experiment with Nagin and Sanders (Nagin et al., 2002), which manipulated monitoring levels at call centers collecting donations for charitable causes. In that study, conducted in a non-union environment, we had access to a direct measure of malfeasance on the part of individual employees as well as direct measures of individual employee perceptions about the employer. The measure of employee malfeasance is the rate at which employees artificially inflated their level of sales in order to earn extra commissions at the expense of the firm. The employer could catch some, but not all, of this activity through costly monitoring of a random sample of calls. When employees worked in centers with very low apparent rates of monitoring, opportunistic behavior increased. However, this increase was observed only for a subset of employees, and, importantly, increased opportunism was most prevalent among workers who had expressed feelings that the employer treated them unfairly, did not care about them, and provided a bad place to work.³³ Employees who perceived themselves to be unfairly treated struck back at the employer (and added to their own income) when the opportunity to do so arose.

Inequality aversion on the part of employees has a number of interesting ancillary predictions about the way employment relationships are organized. Firms that employ people in both low-wage and high-wage occupations must go to some length to be sure that employees in the low-wage occupation do not include the high-wage occupation in their reference group. Failure to make this separation can lead to pressures to either pay employees in the low-wage occupation too much or employees in the high-wage occupation too little. Indeed, it would not be at all surprising to see firms choosing outsourcing to other firms to avoid just these sorts of invidious comparisons.

Similarly, if employees respond to perceived inequities by retaliating along important, but hard-to-monitor dimensions of work effort and quality, firms that engage in highly unequal pay practices ought to seek out ways to reduce the perceived level of inequality. Secrecy regarding pay is a common human resource practice and it obviously makes invidious pay comparisons more difficult. Some companies, such as Walmart and Lincoln Electric, famously go to great lengths to discourage ostentatious executive perks and the depressing effects on morale they might engender.

For publicly traded corporations the compensation of top executives is a matter of public record. In practice, however, these companies adopt complex and opaque compensation practices that make it difficult to understand exactly how much and in what ways top executives are paid. Bebchuk and Fried (2003) argue that anomalous features of executive compensation—such as the reliance on “at the money” stock options rather than stock grants—are best understood as efforts to camouflage pay and so avoid “outrage” from shareholders, employees, regulators and other interested parties.

Levy and Temin (2007) examine these outrage costs from an institutional and historical perspective. They argue that the Federal government enforced a set of informal yet egalitarian social norms on executive pay from the post World War II era through the early 1970’s. These norms were part of a larger set of institutional arrangements that included powerful unions, high minimum wages, and high marginal tax rates for high earners. The actions taken by Reagan administration in the 1980s (notably the firing of the air-traffic controllers) signaled that the Federal government was leaving such decisions as CEO pay strictly to market forces. The degree to which income norms can be shaped by national institutions and economic policy is an important question that would benefit from additional empirical and historical research.

4.2 Effort norms

Fehr and Schmidt’s key idea is that equity concerns constrain firm behavior even in competitive labor markets, because of behavioral features in agency problems inherent in employment relationships. If, as appears to be the case, employees respond to unequal or unfair treatment by taking actions that punish the employer, it is a short further step to presume that morale enhancing activities ought to motivate employees to take actions in the interest of their employer. This is the premise of Akerlof’s (1982) gift exchange model of efficiency wages.

Akerlof’s model plays a central role in behavioral labor economics because it relies on very different sociological and psychological mechanisms than the standard agency approach presented in Section 2 above. Instead of engaging in calculations about the costs and benefits of working harder, employees in Akerlof’s model are motivated by norms governing behavior in the exchange of gifts. If the employer pays employees a wage higher than some reference wage, the employee perceives himself to have received a gift from the employer. This gift creates an obligation to give something valuable in return. In the employment context, the obvious way to reciprocate is to provide the firm with more than the minimally acceptable level of work effort and attention.

Akerlof’s approach to the problem of agency rests critically on the concept of effort norms, i.e., on the idea that individuals are motivated to provide effort in ways that enable them to conform to their self image or social identity. Decent people, so the reasoning might go, return kindness with kindness and so, wishing to preserve the self image of decency, the employee responds to a high wage by returning the favor in the form of high effort to the employer.³⁴

Experimental investigations suggest that reciprocity of the sort identified in Akerlof can survive in competitive environments. For example, Fehr et al. (1998) report results from a laboratory experiment in which sellers have the opportunity to select quality levels above the minimum level enforceable by buyers. In treatments where sellers have the opportunity to do so, they reciprocate high prices with high quality levels. Anticipating this behavior, buyers profit by offering high prices far in excess of the seller’s reservation prices. In treatments where sellers do not have the opportunity to reciprocate, buyers offer lower prices.

If employee effort responds to the perceived “fairness” of wage offers, then policy makers must pay special attention to policies that might shift perceptions of the “fairness” of a wage offer. Policy may be especially likely to affect fairness if individuals care about employer intentions as well as outcomes. A wage of in the absence of a minimum wage might be perceived to be quite fair because employers could have offered a good deal less but chose not to. If, on the other hand, the minimum wage was set to be , then the employer might have to offer a wage above to demonstrate good intentions.

In an important paper, Falk et al. (2006) investigate whether minimum wage laws influence the perceived fairness of wage offers. They set up an experimental labor market in which individual employees (students paid to participate in the experiment) have to decide whether or not to accept a job offered by a firm. Contrary to the conventional self-interest model, but consistent with a fairness-concerns model, individuals had reservation wages significantly above zero. There was also considerable heterogeneity in reservation wages, giving individual firms an upwardly sloping supply curve for labor. Introducing a minimum wage in this labor market had the effect of increasing individual reservation wages considerably—a result consistent with the hypothesis that the perceived intentions of the firm matter in determining fairness. Surprisingly, there appeared to be hysteresis in the effect of minimum wage laws on reservation wages: subjects exposed to the laws after participating in labor markets with no minimum wage laws increased their reservation wages, but subjects who first participated in labor markets with minimum wages did not revise reservation wages downwards when the laws were removed. These results, if they hold outside the laboratory, have implications that extend far beyond the issue of minimum wage laws. Labor market regulations that influence employer scope of action must take into account how these regulations are likely to affect employee perceptions of employer intentions. More provocatively, the hysteresis result also raises the possibility that regulators might not be able to “undo” the effects of policy simply by reversing previous decisions.

In strong form, well-functioning norms can have considerable social value. They can serve to reduce the problems created by agency in many contexts, including employment relationships within firms.

Given their considerable economic value, it is important to understand the social processes that generate and sustain socially valuable effort norms. In a path-breaking economic analysis, Frank (1988) emphasizes the role of emotions in resolving the “commitment problem,” i.e., the problem of eliciting a commitment to constructive cooperation. He argues that rational calculation is often not sufficient to sustain cooperation because by the time the misbehavior occurs, the benefit of punishing the bad actor has often already passed. Emotions, in contrast, can be the foundation of much more powerful sanctions because the commitment to follow through on the action is rooted in the primitive reward structure of the brain. Thus, “cross me and you’ll never work in this town again” is a weak deterrent when uttered by a rational calculator who may decide after the fact that it is not worth the effort to punish the double-crosser. It is a strong deterrent, however, when uttered by someone who gets visceral satisfaction in carrying out his threat regardless of the cost to himself.

From a psychological perspective, emotions emerge from a genetically determined neurological reward system. The triggers of this reward system, however, are shaped by an intense and costly socialization process that trains individuals to have a “conscience,” i.e., to feel strong emotions when they lie, cheat or otherwise disappoint others’ expectations. Evolutionary considerations led Frank to expect that these efforts at socialization will not be entirely successful. Society will be composed of a mixture of types: opportunists who take advantage of chances to free ride and reciprocators who will devote resources to monitoring the behavior of their counterparties and cooperate so long as they perceive others doing the same.³⁵

The idea that populations contain a mixture of opportunists and reciprocators is supported by experiments involving public good contribution games. Fehr and Gaecher (2000) study such games and contrast two treatments. In the “no punishment” treatment, anonymous individuals are randomly allocated to groups of four and are given the opportunity to make contributions towards a public good. Payoffs are such that the dominant strategy is to make no contributions towards the public good. In the “punishment” treatment a second stage is added which gives each individual the opportunity to punish others by subtracting from their payout. Punishing poor contributors is costly, however, and no one interested in maximizing their monetary payoff will choose to punish after the damage is done.³⁶ For this reason one would expect the dominant strategy to be one of “no contribution” in both punishment and no-punishment treatments. This prediction is confirmed in the “no punishment” game: average contributions converge to almost complete free-riding. In contrast, in games with the option to punish in the second stage, individuals do make substantial contributions to the public good and these contributions do not fall over time. Consistent with our effort norms model, subjects are more heavily punished the more his or her contribution falls below the average contribution of other group members. Individuals also exhibit heterogeneous tendencies to free-ride and punish. Depending on the definition, the authors estimate that between 20 and 53 percent of subjects in their study were free riders. Heterogeneity in the tendency to behave opportunistically has important implications for labor markets and personnel practices that we explore further in Section 5.

Emotions can support pro-social behavior in ways other than sustaining irrationally high levels of retaliation against defectors. Ekman (2001), for example, argues that emotional states can be read from the facial expressions of individuals. It follows from this that lying and other opportunistic activities that can elicit strong emotions are harder to sustain during face to face interactions. Valley et al. (1998) investigate this hypothesis in a bargaining experiment which requires negotiators to elicit private information about the true value of an underlying asset when the incentives in the experiment do not support revealing this information truthfully. The study finds that face to face negotiations are more likely to reach mutually beneficial solutions than negotiations conducted over the phone or in writing.

The feelings of guilt and shame that support truth telling and honesty are similar to the emotions that support effort norms, and these emotions are generally thought to be strengthened by physical proximity and face to face interactions (Sally, 2002). A nice laboratory experiment by Falk and Ichino (2006) provides evidence along these lines. In particular, in that study the authors observed “peer effects” in which subjects who would otherwise have provided low effort were motivated to increase effort when physically paired with high-productivity workers. ³⁷

In a remarkable study of cashiers at a national supermarket chain, Mas and Moretti (2009) find that substituting a worker with below average productivity for a worker with above average productivity is associated with an increase in the effort of other workers on the same shift. Low productivity workers are especially responsive to the composition of their co-workers and this peer effect occurs only for low productivity workers who are in the line of vision of the high productivity workers. The effect of high output peers on the productivity of others declines with distance and with the frequency of interaction as measured by the degree to which shifts overlap.

If effort norms indeed require close proximity and frequent interactions within a work group, it is natural to ask whether these motives can operate in large organizations. Very little empirical work has focused on this important issue.³⁸

Effort norms clearly matter within organizations and work groups, and may have important implications for broader labor markets as well. In the gift exchange model, as first set out by Akerlof (1982), the “gift” that results in the optimal reciprocal responses from agents is a wage that exceeds the market-clearing wage. The consequence is equilibrium unemployment (see also Akerlof and Yellen, 1990). If some firms and industries find it important to use gift exchange as a motivating tool, and others do not, then the gift exchange model can be used to explain “dual labor markets,” i.e., to help understand cross-firm and cross-industry wage variation.

As Akerlof and Yellen (1985) and Akerlof et al. (2000) argue, the gift exchange logic—that worker performance depends on a firm’s current wage relative to a reference wage and to the unemployment rate—can be a building block for macroeconomic models. In Akerlof et al. (2000), for instance, a reference wage model is combined with an assumption that some principals adopt “near rational” wage setting rules whereby they ignore the effect of inflation on reference wages when inflation rates are sufficiently low. The consequence is a long-run Phillips curve with the property that a modest rate of inflation (approximately 3% in their calibration) is associated with a lower unemployment rate than is either 0 inflation or high inflation.³⁹

4.3 Professional norms

In professions such as law and medicine, the principal agent problem takes on a special importance. Professionals are in theory the agents of their clients, but professionals enjoy advantages of education, credentials, status and specialized knowledge that make their clients especially vulnerable to exploitation. In order to protect clients from abuse, professions go to great lengths to inculcate norms of professional conduct. This makes professions an especially important venue for analyzing the effect of norms.

In our analysis of physician incentives in a managed care organization with Martin Gaynor (Gaynor et al., 2004), we develop a simple model of professional norms that we adapt here. The model follows the approach used throughout this essay: we modify the agent’s utility function, in this instance to include physicians’ regard for their patients. We posit, in particular, that the socialization of physicians causes them to experience disutility when they adopt a practice style that delivers medical services that are less than the level that the patient would select for themselves (if they were as well informed as the physician). Think of this level of services, , as the level (measured here in dollars) that results when the physician incorporates a patient’s own preferences into his utility function.

We write the utility of a physician treating patients as a function of income earned, , and the deviation of medical services, from the ideal level:

     (29)

where is a convex function that achieves a maximum when is equal to a subjective “best” levels of care, , and the parameters indicate the weight the physician places on each patient’s well-being. Thus physician utility is increasing in both income and services provided when they adopt a practice style with . In a fee-for-service environment where insurers don’t try to “manage” the care physicians provide, one would expect physicians to deliver care at or close to .

Managed care organizations, such as Health Maintenance Organizations (HMOs), often try to influence physician practice styles through the use of financial incentives. The managed care organization we studied, for example, adopted a simple incentive strategy designed to restrict utilization without substantially harming patients: the principal (HMO) offered agents (physicians) a bonus if total annual medical expenditures fell below a target . The probability that a physician’s expenditures on behalf of patients fell below this threshold depended of course on the decisions made on behalf of each patient and also random factors. The probability of earning the bonus, given expenditures and target , is given by the c.d.f. .

Giving the physician responsibility for the allocation of resources across a panel of patients in this way makes sense when physicians have practice norms of the sort characterized in (29). We can illustrate the idea easily with the case in which disutility is . In this case the physician’s best response to a policy, and is found by maximizing

     (30)

This leads to the best response function for the treatment of each patient :

     (31)

The extrinsic reward induces the physician to conserve resources on behalf of the HMO, and if is the same across all patients, say (i.e., there is no favoritism), the physician does so in a sensible way. ⁴⁰ Also, it is easy to confirm that if the second order condition holds, an increase in the bonus induces the physician to reduce expenditures on patients,

     (32)

From (31) and (32) we can see that the intrinsic value the physician places on patients (represented by ) governs the level of expenditures chosen for patients, as well as the power of the extrinsic incentive to alter chosen expenditures.

In our empirical analysis of internal records in an HMO (Gaynor et al., 2004), we found results consistent with the prediction in (32); increased incentive intensity led physicians to reduce expenditures on patients. We also found that physicians cut costs most for outpatient and elective procedures, but not at all for inpatient procedures. Consistent with the model, this suggests that physicians cut costs most where the consequences for patient welfare were lowest.

Two obvious implications of this treatment of professional norms and incentives merit mention. First, patients are not necessarily harmed by incentives that impose constraints on physician actions. Indeed, it is clearly in the interests of patients for their physicians to allocate resources in a reasonable way, because ultimately patients pay for misallocation through higher insurance premiums. Second, physician practice norms of the sort specified above serve to protect patients from potential abuses introduced by cost-containment incentives, especially if the internalization of patient utility is allocated evenly across patients.

Given the pivotal role of professional norms in protecting clients, it is important that attention be paid to the ways in which these norms are established and how they might be undermined. A key example of the latter phenomenon is professional “conflict of interest.” In health care, drug companies famously used gifts and aggressive marketing to influence the prescribing activities of physicians (Avorn, 2004). Conflicts of interest arise in other contexts as well. For example, Jackson (2008) observes that in many financial services markets (including the market for health insurance), brokers who represent one side of the transaction are paid by the other side. These arrangements clearly threaten the ability of brokers to represent the interests of their clients.

Economists have devoted relatively little attention to understanding why practices that create such obvious conflicts of interest persist in markets where principals greatly depend on the independent judgement of professionals. An important exception is a provocative set of articles, Dana and Loewenstein (2003), Moore and Lowenstein (2004) and Cain et al. (2005), which argue that even small gifts can trigger a norm of reciprocity that introduces largely unconscious biases into professional judgements.⁴¹ The fact that these biases are unconscious prevents them from inducing the negative feelings that otherwise cause professionals to conform to norms of acceptable behavior. Laboratory experiments suggest that clients who rely on professional judgements do not adequately adjust their interpretation of professional advice even when they are informed that their agents might be biased. Although the evidence for this view of conflict of interest is far from definitive, the implications for the successful resolution of principal agent models in professional settings are both profound and unsettling.⁴²

4.4 Identity

Our discussion of professional norms focused on the idea that physicians might experience disutility—perhaps profound feelings of discomfort or anxiety—if they deviate from proscribed behaviors with respect to their clients or patients. This approach to economic sociology is discussed at length in the work of Akerlof and Kranton (2000, 2005) on “identity.” Here is the key idea:

The idea that “category” and “situation” can be fundamental elements in preferences enormously expands the range of principal agent models. For instance, to the extent that identity can be manipulated within an organization, identity-based incentives might substitute for extrinsic rewards.⁴³ Just as families and religious communities undertake important and costly investments to ensure that their children internalize a set of values and practices consistent with passing on the family or group’s social identity, so organizations might make investments in practices that persuade employees to adopt goals of the organization, and so mitigate agency problems. These investments are likely to be greatest where financial rewards are most costly to the organization, e.g., when performance measures are especially noisy and where high effort (or high effort at peak times) is critical to the organization’s success. Investment in identity incentives will also be greatest where inculcating identity is cheap, and it is reasonable to suppose that imparting identity is cheapest when agents are young and/or when highly motivated individuals self-select into the organization—an issue to which we return in Section 5.

The great virtue of identity models is that they are highly flexible and therefore able to account for behavior that is anomalous from the perspective of simple agency models. This virtue is a curse, however, when it comes to generating falsifiable hypotheses for testing identity models themselves. One way around this problem might be to focus on a particular relevant social category and seek to understand key norms that can be studied systematically and characterized in a parsimonious way.

A template for this latter approach can be found in a series of careful and nuanced investigations of psychological factors that generate gender differences in economic behaviors. For example, work by Babcock and her co-authors, demonstrates a profound gender-based difference in the inclination to initiate negotiation; “women don’t ask.”⁴⁴ A simple and clear demonstration emerges in an experimental study in which subjects are asked to complete a simple task, and are then put in a position in which there is ambiguity with regard to the payment. In a typical experiment, subjects were told in advance that the payment would be between three and ten dollars. Then, at the conclusion of the session, the experimenter says, “Here’s three dollars. Is three dollars okay?” Eight times as many men as women asked for more money in this experiment. Even in a variant of the experiment in which the experimenter provides cues to signal the social acceptability of negotiation (e.g., with a prompt, “the exact payment is negotiable”), far more men than women take up the opportunity (Small et al., 2007).

In laboratory and field experiments, this disinclination by women to “ask” affects outcomes in negotiated settlements, leading women to do less well than men. Importantly, though, when a woman advocates on behalf of someone else, she is typically more successful than when she negotiates for herself, and indeed is generally more effective than men in this capacity (Bowles et al., 2005). Part of the reluctance to “ask,” it appears, comes from a desire to avoid self promotion.

Along these same lines, Gneezy et al. (2003) find that women respond differently than men to tournament style incentives when these contests involve both men and women. When paid by piece rate or when competing in single sex tournaments, women’s performance is similar to those of men. Niederle and Vesterlund (2007) provide experimental evidence that in comparison to men, women generally shy away from incentives schemes that involve tournament competition.

Gender identity, in short, matters in economically important ways. It is tempting to assert that female identity includes a component that guides women to shy away from competition with men and to reject self promotion. However, it is important to understand that this might not be the whole story, or even the most important part of the story, when using identity to explain gender differences in behavior. There is considerable evidence in psychology that a “kinder, gentler image” is expected of women (to use the expression in Rudman and Glick’s (1999) article on the topic). Women who violate that norm by engaging in self promotion face the potential of backlash, which can entail psychological and material costs (as when a woman is bypassed for promotion because she is seen as “inappropriately assertive”). Thus, even a woman who feels no particular disinclination for self promotion might find it in her self interest to adopt the expected “kinder, gentler” norm (Bowles et al., 2006).⁴⁵

Standard agency models, discussed in Sections 2 and 3, do have implications for gender in labor markets.⁴⁶ The recent work by Babcock, Niederlie, Vesterlund, and their co-authors, discussed above, adds a new and promising perspective for understanding the role of gender in organizations and labor markets. Babcock and Laschever (2003) provide extensive evidence that women’s reluctance to ask often includes an unwillingness to negotiate their own salaries. It follows logically that in labor markets in which there is rent sharing, this psychological phenomenon contributes to male-female wage and income gaps. On the other hand, the title to Niederle and Vesterlund’s (2007) paper—“Do Women Shy Away from Competition? Do Men Compete Too Much?”—suggests an important point: Cooperation, and the willingness to work hard on the behalf of others, are valuable traits, which should receive positive value in the labor markets. An important research agenda going forward is the incorporation of new findings on gender from psychology and behavioral economics into models of organizations and labor market equilibrium for the purpose of investigating these very issues.⁴⁷

There are certainly other important identity categories that deserve attention from behavioral economists who study organizations and labor markets. Ethnicity and sexual orientation are additional identity categories that are important in many contexts, including, quite possibly, the labor market.⁴⁸ Berman’s (2000) economic analysis of ultra-orthodox Jewish groups indicates a strong behavioral impact of religious identity, which induces many Israeli ultra-orthodox men to engage in fulltime yeshiva study into their early 40s, thereby impoverishing themselves and their families.

4.5 Miscommunication and race

Another identity category of indisputable importance is race. There is little theoretical work in economics that explores the role of race in organizational form and compensation practices. A very important exception is Lang’s (1986) “language theory” of statistical discrimination, which focuses squarely on agency and performance within organizations. The starting point of Lang’s analysis is the observation that misunderstanding and misinterpretation are common workplace problems. Lang draws on a wide body of literature in psychology and linguistics to argue that these problems are exacerbated when managers and workers are from different “cultural” or “linguistic” groups.⁴⁹

Following Lang’s lead, Ritter and Taylor (forthcoming) consider a labor market in which there is potential for race-based workplace misunderstandings. Their focus is the possibility that this force contributes to black-white gaps in unemployment. The model of unemployment is the agency-based efficiency wage model outlined in Section 2.2.3.

Suppose that most supervisors in the US are white (perhaps because capital is disproportionately in the hands of whites in the US), and that these managers are more successful evaluating the performance of white employees than black employees. Now recall that in the efficiency wage model set out above, (the standard deviation of the noise) reflects the precision with which managers evaluate workers. The logic of Lang’s arguments leads to the conclusion that is relatively higher when white managers evaluate black workers. If so, the unemployment rate will be higher for blacks than for whites.

To see the logic of the Ritter-Taylor result, recall, from (23), that in an economy with homogenous workers, an efficiency wage strategy of worker motivation leads to the following relationship between the equilibrium wage and unemployment rate :

     (33)

Now suppose that black and white workers are equally productive and thus in equilibrium must be paid the same wage. Suppose also that, as discussed in the last paragraph, there is more noise in the evaluation of black workers than white workers: . Then the following must hold in equilibrium:

     (34)

where is the unemployment rate for black workers and is the unemployment rate for white individuals. Clearly, . We therefore have a potential explanation for racial differences in unemployment rates.

This model is thus consistent with evidence, such as Neal’s (2006), that among men, black-white gaps in the wage are small when one conditions on a measurement of human capital taken when the men were youths (the AFQT), but black-white gaps in unemployment are large. Ritter and Taylor (forthcoming) show that black-white unemployment gaps persist when one conditions on the AFQT, and that unemployment rates are highest for black men who attended high schools in which other students were mostly black. Under the assumption that these men are most likely to find on-the-job interactions with their boss difficult, this evidence is consistent with the model of racial differences in unemployment we have just outlined.⁵⁰

The “miscommunication model of unemployment” outlined above calls attention to a general point that pertains broadly in models of behavioral agency: If the efficient resolution to agency problems is economically important, than labor markets will tend to reward individuals who possess scarce preferences and traits that enhance the effectiveness of firms’ strategies to evaluate, monitor, and provide incentives. Bowles et al. (2001) offer a creative assessment of the labor market returns to such incentive-enhancing traits—traits that might include a low rate of time preference, an intense sense of shame at being without work, perseverance, identification with work goals, and the psychological predisposition to see personal initiative and self determination as important relative to external luck or fate (i.e., to have “internal control” rather than “external control” on Rotter’s “locus of control”).

Bowles et al. (2001) offer an overview of empirical evidence that wages are correlated with measures of some such traits. For example, people with a high degree of internal control earn higher wages.⁵¹ Of course, correlation does not establish causality, and the Rotter measure may simply stand in for on-the-job productivity. Still, the authors persuasively argue that incentive-enhancing traits can matter for labor market outcomes, and may be important for understanding the large amount of unexplained variation typically observed in estimated wage regressions.

A number of recent theoretical papers in behavioral economics explore the implications of heterogeneity in agent traits along some key dimension (identification with the task, degree to which the agent is pro-social, etc.). In Section 5 below we discuss several of these papers. In each case the distribution of traits as taken to be exogenous—a reasonable approach given the goals in each paper. But, of course, in the broader scheme, many of these key traits are shaped by individuals’ home and school environments. This latter point is developed in the seminal work of Bowles and Gintis (1977), who argue that if the education system is to be successful in preparing students for the labor market, the objective function ought to include the development of both cognitive skills and incentive-enhancing behaviors.